pi-openai-toolkit 0.23.1 → 0.24.0

package/src/request-context-cache.ts

@@ -1,155 +1,155 @@
-import type { ResponsesCompatibleRequestPayload } from "./runtime";
-import { LOCAL_WEB_SEARCH_TOOL_NAME, WEB_RUN_TOOL_NAME } from "./web-search/types";
-
-/**
- * Explicit allowlist of fields mirrored from the latest live Responses request.
- * remote_compaction_v2 has no provider payload at session_before_compact time, so
- * the extension reuses only these compact-relevant fields for the same runtime identity.
- * `stream`, `store`, `input`, and trigger semantics are always forced separately.
- */
-export type CompactionRequestExtras = {
-	tools?: unknown[];
-	parallel_tool_calls?: boolean;
-	reasoning?: Record<string, unknown>;
-	service_tier?: string;
-	prompt_cache_key?: string;
-	prompt_cache_retention?: "in_memory" | "24h";
-	prompt_cache_options?: Record<string, unknown>;
-	text?: Record<string, unknown>;
-};
-
-export type RequestContextIdentity = {
-	provider: string;
-	api: string;
-	model: string;
-	baseUrl: string;
-	sessionId?: string;
-};
-
-type CachedRequestContext = {
-	identity: RequestContextIdentity;
-	extras: CompactionRequestExtras;
-};
-
-export type RequestContextCaptureOptions = {
-	/** Standalone Web Search is a local tool and must not enter synthetic compact requests. */
-	excludeWebSearchTools?: boolean;
-};
-
-let cached: CachedRequestContext | undefined;
-
-function isRecord(value: unknown): value is Record<string, unknown> {
-	return !!value && typeof value === "object" && !Array.isArray(value);
-}
-
-function isWebSearchTool(value: unknown): boolean {
-	if (!isRecord(value)) return false;
-	if (value.type === "web_search" || value.type === "web_search_preview") return true;
-	return value.type === "function" && (value.name === LOCAL_WEB_SEARCH_TOOL_NAME || value.name === WEB_RUN_TOOL_NAME);
-}
-
-function sameIdentity(cachedIdentity: RequestContextIdentity, current: RequestContextIdentity): boolean {
-	return (
-		cachedIdentity.provider === current.provider &&
-		cachedIdentity.api === current.api &&
-		cachedIdentity.model === current.model &&
-		cachedIdentity.baseUrl === current.baseUrl &&
-		cachedIdentity.sessionId === current.sessionId
-	);
-}
-
-/**
- * Remember compact-relevant fields from a live Responses request payload.
- * Purely additive metadata: failures are swallowed so caching can never
- * break the provider request path.
- */
-export function rememberRequestContext(
-	payload: ResponsesCompatibleRequestPayload,
-	identity: RequestContextIdentity,
-	options: RequestContextCaptureOptions = {},
-): void {
-	try {
-		if (payload.model !== identity.model) {
-			cached = undefined;
-			return;
-		}
-
-		const extras: CompactionRequestExtras = {};
-		if (Array.isArray(payload.tools)) {
-			const tools = options.excludeWebSearchTools
-				? payload.tools.filter((tool) => !isWebSearchTool(tool))
-				: payload.tools;
-			extras.tools = structuredClone(tools);
-		}
-		if (typeof payload.parallel_tool_calls === "boolean") {
-			extras.parallel_tool_calls = payload.parallel_tool_calls;
-		}
-		if (isRecord(payload.reasoning)) {
-			extras.reasoning = structuredClone(payload.reasoning);
-		}
-		if (typeof payload.service_tier === "string" && payload.service_tier.trim().length > 0) {
-			extras.service_tier = payload.service_tier;
-		}
-		if (typeof payload.prompt_cache_key === "string" && payload.prompt_cache_key.trim().length > 0) {
-			extras.prompt_cache_key = payload.prompt_cache_key;
-		}
-		if (payload.prompt_cache_retention === "in_memory" || payload.prompt_cache_retention === "24h") {
-			extras.prompt_cache_retention = payload.prompt_cache_retention;
-		}
-		if (isRecord(payload.prompt_cache_options)) {
-			extras.prompt_cache_options = structuredClone(payload.prompt_cache_options);
-		}
-		if (isRecord(payload.text)) {
-			extras.text = structuredClone(payload.text);
-		}
-
-		cached = {
-			identity: structuredClone(identity),
-			extras,
-		};
-	} catch {
-		cached = undefined;
-	}
-}
-
-/**
- * Look up the exact live-request identity. When building a synthetic request,
- * supply its producer to omit cache fields unsupported by that model/API.
- * No TTL conversion or implicit cache policy is applied.
- */
-export function getCompactionRequestExtras(
-	identity: RequestContextIdentity,
-	producer?: { api: string; compat?: unknown },
-): CompactionRequestExtras | undefined {
-	if (!cached || !sameIdentity(cached.identity, identity)) {
-		return undefined;
-	}
-
-	try {
-		const extras = structuredClone(cached.extras);
-		if (producer) {
-			const compat = isRecord(producer.compat) ? producer.compat : {};
-			if (producer.api !== "openai-responses") {
-				delete extras.prompt_cache_options;
-				delete extras.prompt_cache_retention;
-			} else {
-				if (compat.supportsExplicitPromptCacheMode === true) {
-					delete extras.prompt_cache_retention;
-				} else {
-					delete extras.prompt_cache_options;
-				}
-				if (compat.supportsLongCacheRetention === false) {
-					if (extras.prompt_cache_options?.ttl !== undefined) delete extras.prompt_cache_options;
-					if (extras.prompt_cache_retention === "24h") delete extras.prompt_cache_retention;
-				}
-			}
-		}
-		return extras;
-	} catch {
-		return undefined;
-	}
-}
-
-export function clearRequestContextCache(): void {
-	cached = undefined;
-}
+import type { ResponsesCompatibleRequestPayload } from "./runtime";
+import { LOCAL_WEB_SEARCH_TOOL_NAME, WEB_RUN_TOOL_NAME } from "./web-search/types";
+
+/**
+ * Explicit allowlist of fields mirrored from the latest live Responses request.
+ * remote_compaction_v2 has no provider payload at session_before_compact time, so
+ * the extension reuses only these compact-relevant fields for the same runtime identity.
+ * `stream`, `store`, `input`, and trigger semantics are always forced separately.
+ */
+export type CompactionRequestExtras = {
+	tools?: unknown[];
+	parallel_tool_calls?: boolean;
+	reasoning?: Record<string, unknown>;
+	service_tier?: string;
+	prompt_cache_key?: string;
+	prompt_cache_retention?: "in_memory" | "24h";
+	prompt_cache_options?: Record<string, unknown>;
+	text?: Record<string, unknown>;
+};
+
+export type RequestContextIdentity = {
+	provider: string;
+	api: string;
+	model: string;
+	baseUrl: string;
+	sessionId?: string;
+};
+
+type CachedRequestContext = {
+	identity: RequestContextIdentity;
+	extras: CompactionRequestExtras;
+};
+
+export type RequestContextCaptureOptions = {
+	/** Standalone Web Search is a local tool and must not enter synthetic compact requests. */
+	excludeWebSearchTools?: boolean;
+};
+
+let cached: CachedRequestContext | undefined;
+
+function isRecord(value: unknown): value is Record<string, unknown> {
+	return !!value && typeof value === "object" && !Array.isArray(value);
+}
+
+function isWebSearchTool(value: unknown): boolean {
+	if (!isRecord(value)) return false;
+	if (value.type === "web_search" || value.type === "web_search_preview") return true;
+	return value.type === "function" && (value.name === LOCAL_WEB_SEARCH_TOOL_NAME || value.name === WEB_RUN_TOOL_NAME);
+}
+
+function sameIdentity(cachedIdentity: RequestContextIdentity, current: RequestContextIdentity): boolean {
+	return (
+		cachedIdentity.provider === current.provider &&
+		cachedIdentity.api === current.api &&
+		cachedIdentity.model === current.model &&
+		cachedIdentity.baseUrl === current.baseUrl &&
+		cachedIdentity.sessionId === current.sessionId
+	);
+}
+
+/**
+ * Remember compact-relevant fields from a live Responses request payload.
+ * Purely additive metadata: failures are swallowed so caching can never
+ * break the provider request path.
+ */
+export function rememberRequestContext(
+	payload: ResponsesCompatibleRequestPayload,
+	identity: RequestContextIdentity,
+	options: RequestContextCaptureOptions = {},
+): void {
+	try {
+		if (payload.model !== identity.model) {
+			cached = undefined;
+			return;
+		}
+
+		const extras: CompactionRequestExtras = {};
+		if (Array.isArray(payload.tools)) {
+			const tools = options.excludeWebSearchTools
+				? payload.tools.filter((tool) => !isWebSearchTool(tool))
+				: payload.tools;
+			extras.tools = structuredClone(tools);
+		}
+		if (typeof payload.parallel_tool_calls === "boolean") {
+			extras.parallel_tool_calls = payload.parallel_tool_calls;
+		}
+		if (isRecord(payload.reasoning)) {
+			extras.reasoning = structuredClone(payload.reasoning);
+		}
+		if (typeof payload.service_tier === "string" && payload.service_tier.trim().length > 0) {
+			extras.service_tier = payload.service_tier;
+		}
+		if (typeof payload.prompt_cache_key === "string" && payload.prompt_cache_key.trim().length > 0) {
+			extras.prompt_cache_key = payload.prompt_cache_key;
+		}
+		if (payload.prompt_cache_retention === "in_memory" || payload.prompt_cache_retention === "24h") {
+			extras.prompt_cache_retention = payload.prompt_cache_retention;
+		}
+		if (isRecord(payload.prompt_cache_options)) {
+			extras.prompt_cache_options = structuredClone(payload.prompt_cache_options);
+		}
+		if (isRecord(payload.text)) {
+			extras.text = structuredClone(payload.text);
+		}
+
+		cached = {
+			identity: structuredClone(identity),
+			extras,
+		};
+	} catch {
+		cached = undefined;
+	}
+}
+
+/**
+ * Look up the exact live-request identity. When building a synthetic request,
+ * supply its producer to omit cache fields unsupported by that model/API.
+ * No TTL conversion or implicit cache policy is applied.
+ */
+export function getCompactionRequestExtras(
+	identity: RequestContextIdentity,
+	producer?: { api: string; compat?: unknown },
+): CompactionRequestExtras | undefined {
+	if (!cached || !sameIdentity(cached.identity, identity)) {
+		return undefined;
+	}
+
+	try {
+		const extras = structuredClone(cached.extras);
+		if (producer) {
+			const compat = isRecord(producer.compat) ? producer.compat : {};
+			if (producer.api !== "openai-responses") {
+				delete extras.prompt_cache_options;
+				delete extras.prompt_cache_retention;
+			} else {
+				if (compat.supportsExplicitPromptCacheMode === true) {
+					delete extras.prompt_cache_retention;
+				} else {
+					delete extras.prompt_cache_options;
+				}
+				if (compat.supportsLongCacheRetention === false) {
+					if (extras.prompt_cache_options?.ttl !== undefined) delete extras.prompt_cache_options;
+					if (extras.prompt_cache_retention === "24h") delete extras.prompt_cache_retention;
+				}
+			}
+		}
+		return extras;
+	} catch {
+		return undefined;
+	}
+}
+
+export function clearRequestContextCache(): void {
+	cached = undefined;
+}